ICA diff in cc1 and cc1plus
dbn.lists at gmail.com
Wed Mar 14 09:43:54 PDT 2007
Manuel brought up a recent regression shown by ICA in cc1 and cc1plus.
I did some investigation on this. I reproduced it running jhalfs, but
it didn't show up in my own scripts. I've got a few DIY tweaks in
there, so I started testing out things. First, I thought maybe
bison/flex had made themselves needed again, but that didn't seem to
be the case.
Then I remembered one other thing Greg recently tweaked for more purity.
One of the things that currently doesn't happen in the chroot
toolchain adjustment for LFS is making gcc prefer the new headers in
/usr/include. If you add '-v' to the sanity check output, you'll see
that it's still looking in /tools/include.
So, I added in the -isystem /usr/include to the gcc specs. Lo and
behold, there was a difference. I don't know what would cause this
difference. Either something with using the new kernel headers or
something with gcc-4.1.x since this was never an issue before when we
continued using the headers in /tools/include.
Now I think I see the issue. We don't apply the glibc upstream patch
in Ch. 5. This patch touches two headers that may be included into
built sources: features.h and libio.h.
I still have to spin a full build to be sure that this does fix the
regression, but I thought I'd throw it out there. Although it may fix
things to patch glibc in Ch. 5, I'd prefer to make gcc use the headers
in /usr/include. That seems like the right thing to do.
More info as it arrives.
More information about the lfs-dev